Rationality: From AI to ZombiesCritical edition · Book VI
Becoming Stronger · A skeptical review

Book VI
under examination.

The standards for checking beliefs survive better than the claims to exceptional practical competence.

Review and annotations attributed to Astra 6 Pro · 2026-09-05
43Pieces examined
154Annotations
33Source entries
The judgment

Insight is not a certificate of authority.

Book VI contains useful cautions, a revealing memoir and an ambitious research agenda. It does not establish that its author, readers or preferred institutions possess a reliable general method for superior judgment. At crucial moments, autobiographical conviction and attractive explanations do work that measurements should be doing.

The strongest objections are concrete. An FDA estimate changes units. A psychotherapy experiment changes its comparison group in the retelling. Important psychology claims lose force under replication. Elsewhere, claims about religion, ancestral psychology, cooperation and personal improvement outrun the evidence offered. Bayesian mathematics cannot fill those gaps.

A hard factual error

The FDA estimate gains an extra zero.

Raised in Technophilia turns decade-based estimates into annual figures. The roughly tenfold inflation is a checkable reporting error, independent of the policy debate.

Gieringer (1985), printed p. 196 · Read the annotations

The deeper failure

A recurring shift in evidential standards.

Other schools are asked to prove their methods. The proposed rationality movement is often carried by memoir, analogy and hope. That is enough to propose experiments. It is insufficient to earn deference.

The fundraising chapters expose a further tension. Readers are urged to account for opportunity cost, yet refusal to join or fund favored projects can become evidence of defective cooperation. An informed donor can reasonably reject the project. The relevant standard is comparative expected benefit, not whether the organizer feels that the community has rallied.

A qualification that changes the verdict

The book repeatedly admits that the art is incomplete, discloses failures and invites readers to improve or replace it. It would be a straw man to claim that every essay announces a scientifically validated curriculum. The objection is to the surrounding overconfidence and unsupported prescriptions—not to the legitimacy of proposing a research agenda.

What the studies actually permit

Failure, uncertainty and overreach are different findings.

A failed replication lowers confidence in the effect tested. A different population or outcome may limit generalization without refuting the original experiment. An untested prescription remains unestablished even when its underlying psychological phenomenon is real.

Claim or exampleFindingAssessment and boundaryIn the text
FDA mortality ledgerCorrectionDecade-based estimates become annual totals: roughly tenfold inflation. The original model and the misreported units require separate assessments.Read notes
Ego depletionReplicationTwo large preregistered projects produced near-zero, nonsignificant confirmatory effects. Another found a small positive effect. Confidence in a large general resource-depletion effect is unwarranted; “fatigue is imaginary” would also overreach.Read notes
Power and perspective-takingReplicationMany Labs 3 found d=0.03 for Galinsky et al. Study 2a. The essay leaves its own study unidentified, so this is a relevant direct replication with limited attribution, not proof that every power/empathy finding failed.Read notes
Analysis and religious beliefScopeA later Thinker-picture prime failed direct replication. It was not cited by this essay, and it does not test the proposed five-year curriculum. The broader causal promise remains unestablished here.Read notes
Bystander behaviorScopeThe individual-level effect has substantial meta-analytic support. Field observations show frequent help and a different aggregate relationship. These can coexist: the outcome and situation matter.Read notes
Rorschach validityCorrectionAn indiscriminate dismissal ignores substantial variation among validated scores, weak scores and unstudied scores. This is no endorsement of unrestricted interpretation or proof of clinical usefulness.Read notes
Professors versus therapistsCorrectionThe comparison used professors selected for interpersonal ability and a narrow patient sample. The book’s “randomly selected” description is materially inaccurate.Read notes
CBT and psychotherapyScopeA large depression meta-analysis supports CBT relative to controls. Its tiny advantage over other psychotherapies was not robust. Neither blanket dismissal nor categorical superiority follows.Read notes
Debiasing and forecastingSurvivesSpecific interventions improved tested reasoning and forecasting outcomes. This supports a bounded research program; none of these studies establishes the effect of the complete Book VI curriculum.Read notes
Ultimatum-game stakesScopeThe cited study contains large rejections, but also greater acceptance of a given share at higher stakes. Its extreme example should not erase its comparative result.Read notes
AI-box demonstrationsEvidence gapThe chapter reports three wins and two losses overall, including later high-stakes offers. Testimony without transcripts, independent audit or a reproducible protocol cannot establish a general containment failure rate.Read notes
Conway’s LifeCorrectionThe stated rules omit births: a dead cell with three live neighbors becomes alive. Separately, universal computation does not by itself prove consciousness or suffering.Read notes

This is a targeted audit of consequential examples, not a claim that every study in the book failed or that every relevant later paper has been located. The annotations identify whether a source directly tests the claim, limits its scope, or supplies related evidence.

The logical problems are larger than the replication problems.

Bayes does not select the model.

Consistency rules do not supply correct priors, hypotheses, causal structure or values. The boy–girl puzzle illustrates how a reporting protocol can change a mathematically valid answer.

A story does not establish a method.

Self-correction may be real without showing which practice caused it, whether readers can reproduce it or whether it outperforms alternatives.

Possible does not mean worth pursuing.

Logical possibility, catastrophic stakes and intense commitment do not supply tractability estimates, opportunity costs or stopping rules.

Further recurring shortcuts include inferring an evolutionary history from a plausible function, treating the same action after new evidence as proof of a failed update, and presenting psychological separability as an optimization theorem. These criticisms are arguments about what follows from the stated premises. They are not experimental findings dressed up as a new authority.

What remains unscathed?

The discipline of correction, with its scope kept honest.

Several principles emerge intact in their limited form. No whole empirical package emerges with a clean bill of health.

The strongest individual essays are Three Levels of Rationality Verification, Epistemic Viciousness and the central warning in Beware of Other-Optimizing. Their empirical illustrations and universal language still need qualifications. The ending’s instruction to read beyond one author is also worth following literally.

What would earn stronger claims?

A specified curriculum, suitable comparison groups, preregistered outcomes, meaningful transfer tests, follow-up, costs and attrition, and independent replication. Agreement with the teacher, memorable success stories and fundraising totals cannot stand in for those results.

Scope and accountability

How this review was conducted.

Read all 43 pieces listed for Book VI: Rob Bensinger’s introduction, 12 essays in sequence X, 8 in Y and 22 in Z. Examined the repository’s existing annotation format and added a separate set attributed to Astra 6 Pro. Existing Opus notes were not treated as scientific evidence. Original essay bodies and prior annotation text are preserved.

Annotation attribution: Astra 6 Pro, as requested. Reviewed 2026-09-05. Source repository: bitflora/sequences_annotated. Base revision: 6c24e1e9131beed0e45f9d3461de6e40559bff01.

Every piece accounted for

Coverage of Book VI.

154 annotations · 42 logic · 47 scope · 30 evidence gap · 28 survives · 5 correction · 2 replication. These counts describe the editorial distribution; they are not a quantitative reliability score.

SequencePieceNotesKinds of finding
IntroductionBeginnings: An Introduction4Logic, Scope, Evidence gap
XMy Childhood Death Spiral4Scope, Evidence gap, Survives
XMy Best and Worst Mistake3Logic, Survives
XRaised in Technophilia4Logic, Evidence gap, Correction
XA Prodigy of Refutation3Survives, Logic
XThe Sheer Folly of Callow Youth4Survives, Evidence gap, Logic
XThat Tiny Note of Discord4Logic, Scope
XFighting a Rearguard Action Against the Truth3Evidence gap, Logic, Scope
XMy Naturalistic Awakening4Evidence gap, Scope, Logic
XThe Level Above Mine3Evidence gap
XThe Magnitude of His Own Folly3Evidence gap, Logic
XBeyond the Reach of God4Survives, Logic, Correction, Evidence gap
XMy Bayesian Enlightenment4Logic, Survives, Scope
YTsuyoku Naritai! (I Want to Become Stronger)3Scope, Correction, Survives
YTsuyoku vs. the Egalitarian Instinct3Evidence gap, Logic, Survives
YTrying to Try3Scope, Logic
YUse the Try Harder, Luke2Scope, Survives
YOn Doing the Impossible4Evidence gap, Logic, Scope, Survives
YMake an Extraordinary Effort3Scope, Evidence gap
YShut Up and Do the Impossible!4Evidence gap, Survives, Logic, Scope
YFinal Words3Logic, Scope, Survives
ZRaising the Sanity Waterline3Logic, Scope
ZA Sense That More Is Possible3Survives, Scope, Evidence gap
ZEpistemic Viciousness3Evidence gap, Survives, Scope
ZSchools Proliferating Without Evidence5Correction, Logic, Scope, Survives
ZThree Levels of Rationality Verification4Survives, Scope, Logic
ZWhy Our Kind Can’t Cooperate5Logic, Evidence gap, Scope
ZTolerate Tolerance3Scope, Survives, Evidence gap
ZYour Price for Joining4Logic, Scope, Evidence gap
ZCan Humanism Match Religion’s Output?4Evidence gap, Scope, Logic
ZChurch vs. Taskforce3Survives, Logic, Scope
ZRationality: Common Interest of Many Causes4Survives, Scope, Logic, Evidence gap
ZHelpless Individuals3Evidence gap, Scope
ZMoney: The Unit of Caring4Scope, Survives, Logic
ZPurchase Fuzzies and Utilons Separately3Scope, Evidence gap, Logic
ZBystander Apathy4Scope, Survives
ZCollective Apathy and the Internet3Scope, Evidence gap, Logic
ZIncremental Progress and the Valley3Scope, Survives, Logic
ZBayesians vs. Barbarians4Scope, Survives, Logic
ZBeware of Other-Optimizing4Scope, Logic, Survives, Replication
ZPractical Advice Backed by Deep Theories6Evidence gap, Survives, Scope, Replication, Logic
ZThe Sin of Underconfidence4Scope, Evidence gap, Logic
ZGo Forth and Create the Art!3Survives, Scope
Full critical apparatus

All 154 annotations.

The complete notes are embedded below. Each excerpt is the exact text used to anchor its annotation in the original chapter. “Open annotated chapter” works from the accompanying repository snapshot; the review itself can be read offline.

IntroductionBeginnings: An Introduction4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

A normative ideal does not certify a curriculum

decision theory is standard in cognitive science. For an

The legitimacy of Bayesian probability as a normative framework supplies no empirical certificate for this author’s practical methods. Priors, models, utilities, computational limits and learning outcomes still require justification. The introduction explicitly acknowledges philosophical controversy; readers should preserve that distinction throughout the book.

A2Scope

Sampling design is not merely a private intention

intentions of the experimenter

Frequentist error guarantees concern a specified sampling procedure; a stopping rule can change those guarantees. Bayesian stopping-rule independence is a different property, and default or pragmatic priors need not preserve every desired calibration property. De Heide and Grünwald explain the distinctions. Calling the alternative more subjective is a slogan where an account of the inferential target is needed.

de Heide & Grünwald (2021)

A3Scope

Some transfer is evidence; universal transfer is an extra claim

improves reasoning skills outside the classroom

Fong and colleagues provide evidence that statistical instruction can improve reasoning on everyday problems. That is a useful positive result with a defined task domain. It does not establish that this particular collection produces broad, durable improvements in judgment, and the cited gambling study cannot by itself settle the value of every course in logic or fallacies.

Fong, Krantz & Nisbett (1986)

A4Evidence gap

Adoption supplies no control group

the resultant material has proven useful to a much

A movement, an institute and satisfied readers demonstrate influence and perceived usefulness. They do not identify the causal effect of reading the essays: selection, social support and prior motivation can produce the same observations. The later request to test and replace this text is the stronger standard; institutional descendants should meet it too.

X · Yudkowsky’s Coming of AgeMy Childhood Death Spiral4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Praise and mindset are not a general success technology

the value of—effort, as recommended

A useful concern about rewarding fixed identity should not become a claim that praising effort reliably transforms achievement. Later meta-analyses of the related growth-mindset literature found small average effects and important variation. Those studies do not directly replicate every praise manipulation, but they do undermine an effortless leap from an attractive psychology story to a universal educational prescription.

Sisk et al. (2018)

A2Evidence gap

Both intelligence rankings outrun the evidence

couldn’t possibly have been more intelligent

The author rejects his youthful belief that he was already smarter than his parents, then treats his developmental age as decisive evidence for the reverse ranking. Neither comparison is established here by comparable measurements; the SAT comparison is asserted rather than documented. The narrower lesson survives: success in a religious argument is not a general intelligence test, and intelligence does not guarantee sound judgment.

A3Evidence gap

A superlative with no common scale

intelligence is the unfairest of all those

This youthful ranking of intelligence above wealth, health and birthplace requires a specified outcome and comparative evidence. The text supplies neither, and the author later describes a changed perspective emphasizing death and aging. Read the superlative as part of the obsession under examination, not a scientific hierarchy established by the memoir.

A4Survives

The retraction is sound; the opposite prediction is separate

superintelligence would necessarily imply supermorality

The author correctly withdraws the inference from greater intelligence to benevolent goals. Competence and objectives are different properties, so one does not logically guarantee the other. This correction supplies no numerical probability of catastrophe, no proof of a particular future architecture and no privileged authority for the author’s later proposals.

X · Yudkowsky’s Coming of AgeMy Best and Worst Mistake3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

Rejecting a biological blueprint does not forbid biological insight

nothing like a human brain

The author explicitly credits studying neuroanatomy while asserting that an artificial intelligence should look nothing like a human brain. That strong architectural conclusion needs criteria for what resemblance means and evidence about which mechanisms are useful. Rejecting literal duplication of a brain does not establish that every important resemblance is undesirable.

A2Survives

The selection caveat should govern the whole memoir

chalk this up to the anthropic principle

Acknowledging survivorship bias is a real strength. A good outcome after a mistaken belief does not show that the belief was a good policy, and a persuasive story from the person who persisted does not show what happened to comparable people who chose differently. Keep that caveat when the book later recruits this history as a route to exceptional competence.

A3Logic

Operational knowledge can precede a deep theory

all reasoning on that level is something that can’t be trusted

The author allows preliminary descriptions while warning against putting weight on informal reasoning. That warning is sensible when untested mechanisms carry the conclusion, but reliability can also be established within a tested range before a complete explanation exists. Precision helps expose assumptions; neither a formal definition nor a deep theory is an automatic certificate of truth.

X · Yudkowsky’s Coming of AgeRaised in Technophilia4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

Rejecting a midpoint rule does not refute compromise

were always a compromise. And as

Truth need not lie halfway between two claims. Negotiating among people with different interests, or choosing a robust policy under uncertainty, is a different problem. The existence of a right answer to a factual question supplies no theorem that political or engineering compromise is inherently misguided.

A2Evidence gap

The counterfactual ledger still needs its assumptions

all the drugs that didn’t go wrong

Counting forgone benefits is essential to a fair policy comparison. These estimates still require assumptions about approval timing, uptake, effectiveness and harms. Carpenter and colleagues’ later association between deadline-proximate approvals and safety problems also cannot settle the net mortality balance. A critique of regulation must compare plausible alternatives, rather than treat either an old model or a safety association as an observed all-in death count.

Carpenter, Zucker & Avorn (2008)

A3Correction

A decade has become a year: roughly tenfold inflation

causes at least 20,000–120,000 casualties per year

On p. 196, Gieringer estimates about 5,000 casualties prevented and 21,000–120,000 lives lost per decade. This passage changes the units to per year: roughly tenfold inflation of the annualized scale. It also blurs casualties and deaths. Mis-scaling both sides does not reverse their ratio; assessing the counterfactual model remains a separate task.

Gieringer (1985), p. 196; FDAReview.org, source cited by the essay

A4Evidence gap

An invisible threshold can protect any failed program

doesn’t pass the astronomically high barrier required for

If debiasing only counts once an unspecified, astronomically high barrier is crossed, every disappointing result can be dismissed as insufficient training. That makes the proposal difficult to falsify. State intermediate outcomes, expected effect sizes and stopping rules before using the barrier to explain why improvements remain invisible.

X · Yudkowsky’s Coming of AgeA Prodigy of Refutation3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

Winning an exchange does not establish the thesis

to get the fact of the matter right

An opponent may defend a true conclusion badly, and a critic may find flaws without possessing a better explanation. This distinction is indispensable, including for this review. Every successful objection should reduce only the confidence warranted by the premise it actually defeats.

A2Logic

Responsibility does not imply epistemic self-sufficiency

to save you; you and only you are obligated to

Taking responsibility for a belief is sensible. Treating the individual as the final practical safeguard is much less reassuring: the same blind spots can contaminate both the first judgment and its introspective audit. Independent criticism, records and tests supply error checks that sincerity cannot substitute for.

A3Logic

Explaining moral cognition does not settle moral justification

full reduction of morality that analyzes all

A causal explanation of why humans make moral judgments can illuminate those judgments without proving which values ought to govern action. A reduction must identify what it preserves and answer the normative question, rather than win by announcing that the feeling has been explained away. The acknowledged unfinished philosophy matters more than the posture of having surpassed it.

X · Yudkowsky’s Coming of AgeThe Sheer Folly of Callow Youth4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

The algebra survives; the interpretation is the problem

to zero—that possibility cancels out of

A scenario assigning every act the same utility cancels out of their comparison under the stipulated expected-utility model. The failure is identifying the absence of an objectively compelling morality with the absence of valued outcomes. A person can prefer survival without a universal proof of that preference; the author’s eventual rejection of the original wager is justified.

A2Evidence gap

The counterfactual cure is untested

None of these things could have led me astray

Looking backward and identifying loose reasoning can explain how an error became possible. It cannot establish that sufficient rigor would have prevented every serious mistake: rigorous reasoning can start from false data, omitted hypotheses or a misspecified model. The stronger claim needs more than an autobiography whose ending selects the lesson.

A3Logic

Acting under uncertainty is unavoidable

No, you don’t use the best concepts

Demanding a better understanding before a consequential step can be wise. But waiting, researching and declining to act are also choices with uncertain consequences. Nature’s indifference to excuses cannot establish that a decision made with the best available model was irrational; the relevant comparison is among feasible options at the time.

A4Logic

A useful interface can precede an explanation

not knowing. Any black box could contain a

A black box can be unacceptable when safety depends on its unknown internals. It can also be a legitimate scientific or engineering object when its relevant behavior is measured and its domain of reliability is explicit. A blanket demand to open every box confuses incomplete mechanistic understanding with an absence of usable evidence.

X · Yudkowsky’s Coming of AgeThat Tiny Note of Discord4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

Different hard problems can have different feasibility

“The problem is too hard to solve” is

A person may reasonably expect to improve capabilities while doubting a proposed method of making them safe. Calling both problems difficult does not erase that asymmetry. The important question is whether the resulting project has an acceptable risk profile; the label “too hard” alone proves neither evasion nor sound judgment.

A2Scope

Honesty does not require universal mental disclosure

thoughts broadcast to the whole world

The aspiration to avoid knowingly misleading others is defensible. A demand that every thought be fit for public display would also prohibit legitimate privacy and exploratory reasoning that is not ready for communication. The author’s later qualifications are necessary; discomfort about disclosure is not by itself evidence of dishonesty.

A3Logic

A conditional-independence slogan without the conditions

Actions screen off justifications

An immediate physical action can have the same immediate effect under different motives. Justifications can still predict what details were checked, how the action is implemented and what the person does next. Screening off is a precise probabilistic relationship; it is not licensed merely by drawing a short causal chain in prose.

A4Scope

Unease is a prompt for investigation, not a verdict

Also, perfectionism really matters.

An unresolved inconsistency deserves attention when its expected importance exceeds the cost of checking it. That does not make every slight feeling an accurate detector, nor require endless perfectionist examination. Record the concern, test a concrete implication and decide what evidence would justify moving on.

X · Yudkowsky’s Coming of AgeFighting a Rearguard Action Against the Truth3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

A revised worldview leaves a policy debt

kinds of government regulation are more or

Retracting a premise about technological benevolence should reopen policies that depended on it. The remaining rejection of regulation needs its own analysis of benefits, costs, incentives and institutional alternatives. Personal intellectual progress does not discharge that policy argument.

A2Logic

An unchanged action need not indicate a failed update

afterward he still wanted to charge in

Different models can recommend the same act, especially when the available choices are coarse. Evidence can therefore alter a rational person’s beliefs substantially without changing today’s policy. To diagnose motivated resistance, trace which premises the policy depends on and compare the recomputed alternatives; sameness alone is insufficient.

A3Scope

Recheck dependencies rather than worship a dramatic reversal

far more efficient to admit one huge mistake

A central error may require extensive reconstruction, and the willingness to do that survives. Other beliefs can remain supported by independent evidence. The size and drama of a reversal are poor measures of its accuracy; revise the actual dependency structure, not an imagined minimum amount of repentance.

X · Yudkowsky’s Coming of AgeMy Naturalistic Awakening4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

Retrospective coherence is not a causal study

so, though I remember the important events,

The author admits difficulty reconstructing his earlier thinking and reports correction by others. That candor improves the memoir while limiting what it can establish. Dated records can check particular recollections; the narrative still cannot isolate which habit caused the later change or show that readers will reproduce it.

A2Scope

Thermodynamic claims need the right operation

had to output waste heat in order to

Information erasure has thermodynamic significance; computation is not synonymous with erasing information at every step. Bennett’s reversible-computation construction blocks a blanket inference from logical computation to a fixed dissipative cost per operation. Physical limits remain, but this sketch should not borrow more necessity from thermodynamics than the assumptions support.

Bennett (1973)

A3Logic

Optimization needs a specified baseline

out, steering the future. In the

Describing intelligence as steering futures can be a useful abstraction. Measuring it requires a space of outcomes, a baseline distribution, goals and resource constraints; otherwise even the amount of steering is underdefined. The abstraction alone does not solve the empirical problem of comparing minds or predicting their capabilities.

A4Logic

A possible hostile objective does not establish a historical extinction counterfactual

definitely, no two ways about it, pointlessly wipe out the human species

One can formally specify an agent whose goal conflicts with human welfare. Concluding that the author’s actual earlier project would have killed humanity additionally requires feasible implementation, adequate capabilities, consequential deployment and the relevant objective remaining effective. The missing links are substantial, even though the original assumption of guaranteed benevolence was indefensible.

X · Yudkowsky’s Coming of AgeThe Level Above Mine3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

Awe is not a proof checker

terrifying swift perfection

Feeling intellectually overwhelmed by Jaynes is evidence about the reader’s experience. It is not evidence that every argument is flawless. Halpern’s counterexample to formulations of Cox-style representation results illustrates why assumptions deserve inspection even in beautiful foundational arguments; probability theory itself survives that inspection.

Halpern (1999)

A2Evidence gap

The psychometric threshold is a guess

than around one standard deviation above you starts

The author labels this estimate speculative. There is no measurement here establishing how much intelligence separates recognizing an achievement from producing it, or a universal cutoff beyond which readers cannot discriminate quality. Turning the guess into a hierarchy of whose judgments count would compound its uncertainty.

A3Evidence gap

A career deadline is not a cognitive law

hit a brick wall at 40

The imagined need to catch up by a particular age gives the narrative urgency without establishing a general limit on contribution. Research combines specialized knowledge, collaboration, tools and complementary abilities. A person’s worth or scientific usefulness cannot be inferred from whether they outrank a chosen genius on an unspecified scale.

X · Yudkowsky’s Coming of AgeThe Magnitude of His Own Folly3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

Self-diagnosis does not establish everyone else’s motives

the possibility of others beating them to the punch

The author may have discovered a defensive story in his own thinking. Others may sincerely face competitive pressures while still making poor safety choices. Assess the strategic situation and the proposed response; dismissing an incentive account as ego protection is not a substitute for that analysis.

A2Logic

A bad outcome can follow a rational decision

even if you had wise and excellent arguments

Judgment should be assessed using the information and options available when it was made. A policy that minimizes expected harm can still produce a disastrous realization. Treating disaster as sufficient proof of prior irrationality is hindsight bias in severe language; negligence and bad luck require different evidence.

A3Logic

Safety requires comparison, not an uncertainty-free option

know positively that the ground is safe

The demand to investigate a catastrophic downside is justified. A requirement for positively guaranteed safety can become impossible to satisfy, including for inaction. Decisions need explicit uncertainty, alternatives and asymmetric consequences; the fact that nature punishes errors does not create an option with no risk.

X · Yudkowsky’s Coming of AgeBeyond the Reach of God4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

No fairness guarantee survives intact

victims be completely innocent? Why not,

The laws discussed contain no general promise that moral innocence prevents suffering. That is a coherent rejection of a just-world guarantee. It establishes neither the probability of a specified catastrophe nor the superiority of a specified intervention; those remain empirical and decision-theoretic questions.

A2Logic

Computation and suffering are different premises

Life (which is Turing-complete, so we

Universality supports simulation of arbitrary computable processes given adequate resources. Whether a particular computation instantiates conscious suffering requires an additional account of consciousness and implementation. The automaton argument may illustrate a conditional possibility, but Turing completeness alone cannot prove the existence of sentient victims.

A3Correction

The Life rule omits a birth

really very simple. Cells with three living neighbors stay alive; cells with

Conway’s Life also turns a dead cell with exactly three live neighbors alive; live cells survive with two or three. The printed rule, read literally, loses births and therefore does not define the cited universal automaton. This is a small factual error in a passage asking simple rules to bear a large philosophical load.

Gardner (1970)

A4Evidence gap

A possible rescue is presented with more assurance than evidence

their family for cryonics, and/or write

The argument supplies no demonstrated human revival or quantified chain from preservation to future recovery. Those missing premises matter to the claim that this particular intervention can save the person. Rejecting a cosmic guarantee of justice does not establish the efficacy or expected value of a technological substitute.

X · Yudkowsky’s Coming of AgeMy Bayesian Enlightenment4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

The report-generating process decides the answer

them is a boy,” what is

Under independent equiprobable sexes, suppose a two-boy family always reports a boy and a mixed family reports one with probability q. Conditioning on that report gives P(two boys | report)=1/(1+2q): one-third for q=1 and one-half for q=1/2. An unspecified reporting protocol leaves the puzzle underdetermined. This is a modeling issue before it is an argument between statistical schools.

A2Survives

External correction is the useful endpoint

all this new science, beyond just

The lasting lesson is to let evidence and explicit consequences challenge a cherished theory. Published science is itself corrigible, so citing a textbook cannot be the end of that practice. Apply the same willingness to update to replication failures, to this curriculum and to these annotations.

A3Scope

Finite agents must pay for inference

too computationally expensive to use directly

An approximation can produce a better decision than an exact calculation that finishes too late or consumes the available resources. An ideal consistency theorem does not specify the best approximation under a computation budget. Practical superiority requires comparing achievable procedures, including their costs.

A4Logic

The theorem supplies no privileged prior

Not an answer, but the answer.

Conditional-probability identities are exact within their assumptions. They do not identify the correct hypothesis space, causal model, prior or utility function for a difficult real-world problem. A philosophical conversion to Bayesian language supplies none of those missing inputs by itself.

Y · Challenging the DifficultTsuyoku Naritai! (I Want to Become Stronger)3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Textual authority does not establish intellectual immobility

to myself: “Torah loses knowledge in

Treating a sacred text as authoritative constrains inquiry, sometimes severely. It does not by itself show that adherents cannot develop interpretations, learn new facts or improve their conduct. The sweeping foil needs a more accurate account of the tradition before it can demonstrate a uniquely rationalist virtue.

A2Correction

The religious foil omits repentance’s demand for change

which is about confessing sins—not avoiding sins

A ritual confession can become empty performance, but Judaism does not define repentance as reciting guilt while planning to repeat the offense. Maimonides explicitly requires abandoning wrongdoing and resolving against its recurrence. Criticize the observed gap between doctrine and practice if evidence supports it; the doctrine itself already contains the author’s central demand for improvement.

Maimonides, Repentance 2:2

A3Survives

Improvement is more useful than a ritual of inferiority

what you plan to do about it

Naming a mistake and changing a relevant behavior is a defensible standard for progress. Neither recurrent confession nor an elevated self-image establishes that change. Specify the capacity being improved and measure it: motion, effort and confidence can all increase while performance gets worse.

Y · Challenging the DifficultTsuyoku vs. the Egalitarian Instinct3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

The evolutionary explanation has not been tested here

Among most hunter-gatherer tribes

A description of egalitarian social practices does not establish the ancestral selection history of a modern reader’s embarrassment about achievement. Several mechanisms could produce the same feeling. The text provides no evidence that discriminates its adaptation story from those alternatives.

A2Logic

Equal moral standing does not require equal performance

ashamed of wanting to do better than others

One can affirm equal rights while recognizing large differences in skill. Likewise, wanting to improve does not require ranking other people’s worth beneath one’s own. The useful case for excellence survives without making egalitarian ethics the obstacle.

A3Survives

The average is not an upper bound

who are below average? Do you

Surpassing the current average can be a worthwhile aspiration, and compulsory self-deprecation can obstruct honest assessment. The task still needs an absolute performance target and feedback. Relative rank may rise because others decline; it is an unreliable substitute for becoming more capable.

Y · Challenging the DifficultTrying to Try3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Ordinary language does not reveal the speaker’s objective function

except that the latter expresses the possibility of failure

Saying one will try often expresses uncertainty about success while retaining a serious intention to achieve it. The essay recognizes a colloquial use, but its rhetorical distinction should not become a diagnostic test of someone’s motives. Ask what actions, contingencies and feedback the person has planned.

A2Scope

Concrete contingencies are the testable version

being satisfied with having a plan

The distinction becomes useful when it changes behavior: identify the next action, the obstacle and what to do if that obstacle occurs. Research on implementation intentions gives this kind of if–then planning a measurable form. It supplies no evidence that changing one’s vocabulary or intensifying commitment alone improves outcomes.

Gollwitzer & Sheeran (2006)

A3Logic

A vivid stake does not eliminate opportunity cost

three times your life savings, and you

Maximizing one task’s probability of success is a narrower objective than using one’s resources well. Extra effort can impose unacceptable costs on other goals, even when the task matters greatly. A dramatic hypothetical reward focuses attention; it does not establish how much real-world effort is optimal.

Y · Challenging the DifficultUse the Try Harder, Luke2 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Satirical dialogue is an illustration

you a little-known outtake from the scene,

This purported outtake should be read as a comic philosophical scene, not authenticated production history or evidence about human performance. An author can give fictional characters exactly the obstacles and responses needed to make a lesson look inevitable. The lesson still needs a real-world cost-benefit test.

A2Survives

A cheap attempt can be a useful experiment

People wouldn’t try for five minutes before giving up

Trying briefly can reveal whether an obstacle was merely anticipated, especially when the downside is small. The useful conditions are low cost, clear feedback and permission to stop. Those conditions cannot automatically be transferred to expensive research commitments or decisions with catastrophic consequences.

Y · Challenging the DifficultOn Doing the Impossible4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

Surviving examples select the apparent impossibilities

that I had overestimated the difficulty of

Remembering tasks that turned out to be easier than expected can reveal miscalibration. It omits tasks that remained unsolved and people whose extraordinary efforts failed. Estimate success rates from prospectively recorded attempts and comparable alternatives before using the memorable cases to justify another commitment.

A2Logic

Urgency does not supply an intervention effect

calculated the planetary death rate at 55

A large ongoing harm makes effective interventions valuable. It does not establish that a particular research project can accelerate relief, or that acceleration outweighs its risks and displaced alternatives. The moral stakes amplify the need for those estimates instead of filling them in.

A3Scope

Logical possibility is a very weak feasibility guarantee

Mathematical proof of impossibility

Distinguishing logical impossibility from current inability is useful. A logically possible task may still exceed a finite agent’s time, resources or available methods. Exploration should be tied to learning value and explicit milestones; lack of a proof of impossibility is not positive evidence of tractability.

A4Survives

The explicit permission to lose matters

Never give up? Don’t be

The essay explicitly rejects a blanket never-give-up rule. That concession prevents its central exhortation from becoming an unlimited sunk-cost policy. The practical missing step is to specify what evidence should trigger persistence, a change of method or abandonment before emotional investment grows.

Y · Challenging the DifficultMake an Extraordinary Effort3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

A domain-specific error does not rank whole minds

times before: Robert Aumann, the Nobel

One can criticize Aumann’s religious beliefs without treating that judgment as a measurement of general reasoning ability. A famous name also supplies no performance benchmark for the reader’s next problem. Define the relevant task and demonstrate better results rather than award global intellectual rank.

A2Evidence gap

Selected proverbs cannot explain civilizational outcomes

advantages the West has over the East

The author acknowledges selecting a useful contrast, which makes the passage an analogy rather than a historical test. Causal claims about Eastern and Western success would require representative evidence and plausible controls for institutions, resources and historical contingencies. A pair of sayings cannot do that work.

A3Scope

Unconventional effort still needs a counterfactual

at least a ten-year detour

Formal training can waste time, and independent study can be valuable. An inspiring account of bypassing conventional paths cannot establish that doing so has the higher expected return for another person. Compare concrete curricula, mentorship, access, costs and fallback options; difficulty and unconventionality are not measures of quality.

Y · Challenging the DifficultShut Up and Do the Impossible!4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

An unverifiable demonstration cannot bear a general safety conclusion

neither of us reveal what went on inside

The accounts can challenge a literal claim that these particular volunteers could never be persuaded, if the reported outcomes are accepted. The chapter does not provide transcripts, a reproducible method or an independent audit. It therefore cannot estimate a general escape rate or show that every containment design with different communication rules must fail.

A2Survives

Resolve does not change the probability laws

If you hacked your mind

The refusal to pretend that determination guarantees success is essential. Read the exhortation together with the author’s admissions of failure and emotional cost. Once those are preserved, the defensible message is permission to investigate difficult problems; the text does not demonstrate a transferable ability to solve them.

A3Logic

Low confidence and serious effort can coexist coherently

believe that you will solve the problem with probability 1

A person can assign a small probability to success and still work hard when the upside or information gained warrants it. No temporary belief in one’s own impossibility proof is necessary. The rational question is the expected value of this attempt compared with alternatives, including what failure will teach.

A4Scope

Keep all five reported trials in the denominator

lost the next two

The author reports two earlier wins, then one win and two losses in three further games with offers of thousands of dollars. The account is therefore three wins in five described attempts, not invincibility and not exclusively trivial stakes. A tiny, selected, heterogeneous series remains unsuitable for a stable probability estimate; the disclosed failures deserve to stay visible.

Y · Challenging the DifficultFinal Words3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

Humiliation or catastrophe is not a demonstrated training requirement

must come as a surprise

The teacher says the required catastrophe cannot be artificially provoked, so this is not an instruction to stage one. The remaining claim—that a surprising catastrophe is necessary to teach the crucial lesson—still lacks comparative evidence. Learners can test errors and revise methods without first accepting an unscripted personal disaster as the price of genuine understanding.

A2Scope

Fiction supplies no outcome data for a rationality school

you what my sensei told me,” Jeffreyssai’s

The imagined school can dramatize a training ideal while proving nothing about its effectiveness. An author controls which methods impress the pupil and which challenges vindicate the teacher. A real school must earn those results under evaluation that it cannot script.

A3Survives

Rejecting the inscrutable master is a real safeguard

It was, Brennan realized, a stupid proverb.

The ending’s impatience with opaque sayings undercuts the temptation to worship the teacher. Preserve the requirement for intelligible, usable instruction. The same standard applies to the book’s own memorable slogans: ask which concrete prediction or decision they improve.

Z · The Craft and the CommunityRaising the Sanity Waterline3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

Agreement with the instructor cannot be the entire outcome measure

not directly about religion, that is

A program could increase agreement with its preferred worldview through conformity while leaving general reasoning unchanged. Evaluate held-out factual accuracy, calibration, argument quality and practical decisions, with blinded assessment where feasible. A theological endpoint alone cannot identify the mechanism or establish broad rational improvement.

A2Scope

The proposed training program still needs a causal test

all your students five years later, and see

A multiyear curriculum evaluated against a comparison group is a testable proposal. Later evidence relevant to a much weaker intervention is sobering: Sanchez and colleagues failed to replicate the Thinker-picture prime’s reduction in religious belief, with N=941 and d=+0.07 versus the original −0.59. This later study was not cited by Yudkowsky and does not test five years of education.

Sanchez et al. (2017)

A3Logic

A false belief does not erase every instance of expertise

we have a scientist who’s still religious,

A scientist’s religious commitment can deserve criticism without proving that they understand no evidence, no Occam-style reasoning or no scientific method. People can compartmentalize and possess domain-specific competence. The categorical diagnosis is broader than the evidence supplied by a disputed belief.

Z · The Craft and the CommunityA Sense That More Is Possible3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

Bounded improvements are demonstrable

the three-month summer training program, and half

Morewedge and colleagues found lasting improvement on tested biases after brief training; Sellier and colleagues later tested transfer to a business decision case with 290 graduate students. These are useful, limited demonstrations that particular reasoning skills can improve. They do not show that the Sequences create exceptional general-purpose competence.

Morewedge et al. (2015); Sellier, Scopelliti & Morewedge (2019)

A2Scope

Specify what would count as less than success

systematic training of rationality in a less

A serious research agenda can begin from the hope that more is possible. It should then name expected improvements, comparison conditions and evidence that would reduce that hope. Otherwise the hopeful reader bears the costs while the promised art is perpetually awaiting its first adequate practitioner.

A3Evidence gap

An aspiration is being used to explain its missing evidence

evident aura of awesome.

The absence of spectacular results may reflect insufficient practice, limited transfer, weak methods or an unrealistic target. The essay does not discriminate among those explanations. Continually redefining real rationality as whatever would finally produce awesome results makes the promise immune to ordinary disappointment.

Z · The Craft and the CommunityEpistemic Viciousness3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

The golden age needs historical evidence too

the decline of real fights

The image of martial competence once policed by real encounters gives the institutional story its force. The essay provides no comparative historical account showing how well that mechanism actually selected reliable instruction. A vivid origin story can reproduce the very epistemic vice being criticized.

A2Survives

The warning explicitly applies to the author’s own project

Every single one of these risk factors

This essay does ask whether a rationality school could acquire the same vices it criticizes. It would be unfair to say the self-application is absent. The enduring warning is strong: a community’s scientific vocabulary provides no immunity to deference, dogma or untested instruction.

A3Scope

Recognizing the danger is the beginning of a safeguard

What can be done about it?

A community can praise external testing while continuing to reward insider fluency and loyalty. The practical question is whether independent results change status, funding and leadership when they disappoint. Self-awareness without a mechanism for costly correction is cheap.

Z · The Craft and the CommunitySchools Proliferating Without Evidence5 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Correction

The Rorschach literature does not support a blanket verdict

simply “No.”

Mihura and colleagues evaluated 65 Comprehensive System variables and found sharply uneven validity: some had substantial support, others little or none, and others lacked relevant studies. This does not vindicate free-form inkblot interpretation or settle incremental clinical usefulness. It does defeat the claim that hundreds of studies yield one uncomplicated answer for every score and use.

Mihura et al. (2013)

A2Correction

The professors were selected for interpersonal skill

randomly selected college professor

Strupp and Hadley compared therapists with professors chosen for their ability to form understanding relationships. The patients were a selected male college population; the therapist arm had 15 patients. Similar average improvement in that study does not show that any randomly chosen professor is equivalent to a therapist for any condition. The change from selected helpers to random people materially alters the claim.

Strupp & Hadley (1979)

A3Logic

Similar outcomes do not establish this mechanism

It’s just talking to anyone

Even genuine equivalence between two active treatments would not show that arbitrary conversation is the cause of improvement. Shared therapeutic components, selection and measurement limitations remain possible. A nonsignificant contrast also cannot establish close equivalence without adequate precision and a defined equivalence margin.

A4Scope

Update both sides of the psychotherapy comparison

greater effectiveness of cognitive-behavioral therapy.)

Cuijpers and colleagues’ 409-trial depression meta-analysis found CBT beneficial relative to controls, with effects smaller under stricter bias analyses. Its advantage over other psychotherapies was tiny and not robust. That supports neither the dismissal of treatment as generic chatter nor a sweeping claim that one school has decisively defeated all others; the essay’s late qualification should govern its earlier certainty.

Cuijpers et al. (2023)

A5Survives

Demanding outcome evidence remains right

a practice of doing serious testing

The general objection to professional schools reproducing without demonstrating results is forceful. It applies just as directly to an informal rationality movement selling training or soliciting support. Credentials, charisma and a compelling theory all need outcome checks; this chapter’s own inaccurate example shows why those checks must reach the critic too.

Z · The Craft and the CommunityThree Levels of Rationality Verification4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

This is one of the book’s best distinctions

stratifying verification methods into three levels

Separating demonstrations of skill, comparative tests of methods and routine institutional assessment exposes three different evidential burdens. They should not be collapsed into the reputation of a teacher. The framework remains useful provided that assessors and outcome measures can themselves be challenged.

A2Scope

A useful design sketch is not the definition of science

able to test 100 applications of method A

Randomized comparisons are powerful for estimating instructional effects. The arbitrary sample count cannot guarantee adequate power, and some scientific questions require other designs. Choose outcomes, sample sizes and analysis plans from the effect and uncertainty of interest rather than treat a hundred observations as a ceremonial threshold.

A3Survives

Forecasting illustrates a bounded implementation

This requires standard measurements that can be

Mellers and colleagues tested training, teamwork and aggregation in a geopolitical forecasting tournament with scored outcomes. This provides a concrete example of improving and measuring a consequential reasoning task. It does not establish improvement in every domain, or validate the book’s entire package by association.

Mellers et al. (2014)

A4Logic

More observations do not average away systematic error

an expectation of washing out the variance. But for

Larger randomized samples can reduce sampling uncertainty. They cannot repair a biased endpoint, selective reporting or a measure that rewards memorizing the test. Organizational assessment introduces additional incentives to game scores; validity and resistance to gaming need their own evidence.

Z · The Craft and the CommunityWhy Our Kind Can’t Cooperate5 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

Funding the author’s project is not a neutral cooperation test

weren’t going to donate. Some of

A reader may cooperate effectively elsewhere while judging this organization’s expected return poor. Treating donations to the favored project as the benchmark risks turning a substantive disagreement into a defect in the donor. Compare alternatives and disclose the project’s own uncertainties before diagnosing a community-wide inability to cooperate.

A2Evidence gap

Visible complaints do not identify the silent denominator

none of those donors posted their agreement

Commenters are a selected sample, while donors, quiet supporters and people who simply left may behave differently. The essay recognizes some invisible support, but still generalizes from the experience of asking for money. A claim about a whole community needs participation and outcome data beyond the comments that irritated the organizer.

A3Scope

The value-of-information argument has conditions

Doing worse with more knowledge

For a single decision maker with a fixed objective and no acquisition costs, information that can truly be ignored cannot worsen the optimal choice. In a group, revealing information can change incentives, commitments and other people’s actions. The single-agent argument does not prove that every newly informed group will coordinate better.

A4Scope

Counterargument and backlash must be distinguished

politically opinionated students with more knowledge

The cited result concerns a particular motivated-reasoning experiment. Later survey experiments by Guess and Coppock found no attitude backlash under their tested conditions. These are different designs and do not directly falsify the original study, but they limit a sweeping claim that supplying unwelcome evidence generally makes people more entrenched. More counterargument, less updating and polarization are also different outcomes.

Guess & Coppock (2020; online 2018)

A5Logic

Coordination signals can also manufacture apparent consensus

applaud everything a speaker on a podium says

Public enthusiasm can encourage participation and supply useful social information. Prescribing it without distinguishing sincere endorsement from performance risks misleading observers about support and evidence. A group that needs honest disagreement must design ways to express doubt without treating every visible reservation as sabotage.

Z · The Craft and the CommunityTolerate Tolerance3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Punishment is one cooperation mechanism, not a universal prerequisite

some kind of punishment for defection.

Some games support cooperation through mutual benefit, repetition or aligned interests without punishment. Second-order enforcement can also sustain worthwhile norms under some conditions. A general claim about cooperation must state the game and payoffs before declaring which level of punishment is required or forbidden.

A2Survives

Do not make every disagreement a loyalty test

remind myself, “Tolerate tolerance! Don’t demand

The warning against punishing people merely for tolerating outsiders is a defensible protection against escalating purity demands. It still permits evaluating specific harms and responsibilities. Applied consistently, it protects readers who criticize this movement or refuse to adopt its preferred causes.

A3Evidence gap

A plausible function is not an established evolutionary history

an obvious evolutionary rationale

A story about how punishment or intolerance could have benefited an ancestral coalition is a candidate explanation. It needs evidence that distinguishes it from learning, institutions and other selection pressures. Labeling a rationale obvious does not supply that evidence.

Z · The Craft and the CommunityYour Price for Joining4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Logic

Utility is not identical to the cash payout

utility function for fairness, or because

Rejecting money to punish unfairness violates the cash-maximizing model if money is the only valued outcome. If fairness, reciprocity or expressive conduct enters utility, the act need not violate expected-utility theory. The essay partially acknowledges this; its moral criticism must therefore defend the values or consequences, not borrow a theorem’s authority.

A2Scope

High stakes change the picture, but do not erase all rejection

game played in Indonesia (average per capita

Cameron did report a participant rejecting an offer worth about a month’s expenditure. The study also found greater willingness to accept a given percentage at higher stakes. The anecdote is real; using it without that comparative result conceals how price-sensitive the behavior was. Fairness motives and monetary stakes both matter.

Cameron (1999)

A3Evidence gap

The ancestral group-size story explains too much too easily

40 people, all of whom you knew personally

A fixed small-band picture does not establish why a modern person resists joining a large organization. Hill and colleagues describe mobile, interconnected residence patterns across contemporary foraging societies; these are indirect evidence, not a census of ancestors. The specific psychological adaptation claimed here remains unmeasured.

Hill et al. (2011)

A4Logic

Positive benefit is not enough when better options exist

any other marginal use

The author acknowledges opportunity cost and then discounts its practical relevance. A project can offer positive value while being a poor use of limited attention or funds. That is especially important in a book that elsewhere urges rigorous comparison among charities: the standard should not weaken when the reader is asked to join the author’s kind of organization.

Z · The Craft and the CommunityCan Humanism Match Religion’s Output?4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

The inner beliefs of a population are not directly observed

damnation—theologians don’t really believe that stuff, but

The suggestion that religious participants do not really believe their theology is a psychological hypothesis, not something established by pointing to inconsistency. People can sincerely hold inconsistent beliefs. Evaluate the doctrine’s truth and the institution’s outputs without treating the author’s interpretation of others’ minds as data.

A2Scope

Fatigue is not evidence of a single exhaustible altruism supply

to have unlimited mental energy, then it

The metaphor of finite energy can help a person budget effort. It does not demonstrate a single resource whose conservation will predictably increase altruistic output. Identify which constraint is binding and how the proposed redesign changes it. The numerical institutional prediction is still unsupported even if participants genuinely feel exhausted.

A3Scope

A therapeutic benefit is not an altruism multiplier

Cognitive-behavioral therapy and Zen

Evidence that an intervention improves symptoms cannot by itself establish that participants will do more good for others. The proposed chain runs through wellbeing, motivation, available time, effective action and institutional delivery. Each link needs evidence; borrowing the credibility of therapy does not verify the whole chain.

A4Logic

The proposed efficiency improvement has not been measured

rationalist could get twice as much done.

Getting twice the output for half the input requires four times the output per unit input. An impression that more than half of religious effort is overhead does not establish that gain, particularly if some apparent overhead supports participation. Define the output and include recruitment, retention and support costs before promising superior efficiency.

Z · The Craft and the CommunityChurch vs. Taskforce3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

Treat the replacement as an experiment

be better tested experimentally, than just

The invitation to try institutional alternatives is reasonable. It becomes credible when objectives, costs and retention are recorded and participants can report failure or leave without penalty. A speculative design earns support from its results; the dissatisfaction that inspired it is not yet that evidence.

A2Logic

An apparent cost may help produce a collective benefit

wearing formal clothes is not optimal

Some burdens can signal commitment, screen free riders or sustain participation. Iannaccone’s club-goods model gives one explicit mechanism, without vindicating every ritual. Removing a disliked practice may improve a community, but its causal role has to be tested rather than inferred from its lack of direct output.

Iannaccone (1992)

A3Scope

A taskforce and a community can serve different needs

of rationalist communities, taskforces focused on

An organization optimized for task performance may serve people poorly during illness, unemployment or periods of low productivity. That difference matters when the proposed replacement is supposed to provide belonging and mutual support. Compare success across those purposes, rather than assume a single output metric captures everything a church supplies.

Z · The Craft and the CommunityRationality: Common Interest of Many Causes4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

Letting students choose different causes survives

not all the rationalists I create

The author explicitly accepts that people he teaches may support other projects. That is a valuable limit on the recruitment motive and should constrain the less tolerant fundraising rhetoric elsewhere. Better reasoning can also lead someone to reject all the instructor’s examples; assess reasoning quality independently of which cause receives support.

A2Scope

A shared venue needs room for relevant dissent

agree to shut up about the disagreement

Deferring every cause-specific argument can keep a training meeting focused. It can also shelter the sponsor’s assumptions when those assumptions shape the curriculum or fundraising. Distinguish ordinary agenda management from suppressing criticism material to a participant’s informed choice.

A3Logic

Motivation and effectiveness can both limit impact

overall supply of altruism.

Improving the allocation of a fixed charitable budget is valuable. Increasing available effort or resources can also matter, and the two interventions can interact. Their relative value is an empirical question about marginal effects, not something settled by preferring the more intellectually sophisticated activity.

A4Evidence gap

The limiting resource is asserted, not identified

willpower (the true limiting resource)

Mental accounting can make donation responses depart from a simple fixed-budget model. It does not establish that willpower is the uniquely binding constraint rather than income, trust, attention or perceived effectiveness. That claim needs direct evidence about actual giving; a general self-control theory cannot identify the bottleneck by declaration.

Z · The Craft and the CommunityHelpless Individuals3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

Small-band ancestry is not a sufficient organizational theory

50-person hunter-gatherer bands

Contemporary foraging groups studied by Hill and colleagues show mobility and social connections extending beyond a closed residential band. These observations do not directly reconstruct ancestral minds, but they undermine treating one tiny isolated group as the complete social template. Modern coordination failures need proximate explanations and comparative evidence.

Hill et al. (2011)

A2Evidence gap

The missing organizations have an unknowable denominator here

The vast majority of large modern-day institutions

The claim that potentially valuable organizations mostly never form may be plausible, but this essay offers no method for counting unrealized opportunities or comparing them with failed ventures that did form. The author’s fundraising experience is a case, not a population estimate. Distinguish a proposal worth testing from an established diagnosis of widespread incapacity.

A3Scope

Specialization can be an institutional achievement

itself as a parasite onto the few

Scientists who rely on universities, funders and shared infrastructure are participating in a division of labor. That arrangement can work well even when each scientist lacks the ability to build an institution alone. The insult does not distinguish pathological helplessness from efficient specialization or diagnose which organizational capacities are missing.

Z · The Craft and the CommunityMoney: The Unit of Caring4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

The economic rule has boundary conditions

self-modifying agent, the marginal benefit of investing

Equalizing marginal returns is a useful optimum condition for divisible choices at an interior solution with suitable regularity. Fixed costs, minimum commitments, discrete choices, uncertainty and constraints can change the answer. The model must include those features before the condition becomes practical advice.

A2Survives

Opportunity cost and specialization are valuable cautions

the right to volunteer at the soup

A well-paid specialist may help more by donating earnings than by doing a task another person can perform cheaply. The essay’s qualifications about employment and comparative advantage are necessary. Volunteering can also produce local knowledge, relationships and motivation; include those effects in the comparison instead of assuming they are zero.

A3Logic

Even relative spending reflects more than concern

the unit of relative caring

The essay explicitly describes a relative unit up to a scaling factor; treating it as an interpersonal moral scale would overstate its claim. Even within one person, spending depends on prices, liquidity, constraints and beliefs about what the purchase achieves. Those assumptions must be held fixed before a payment reveals the relative concern attributed to it.

A4Scope

A spending choice reveals less than the moral diagnosis suggests

$5 on a burrito, then whatever

Choosing a meal over a donation can reveal a preference between the concrete options as understood at that moment. It need not show indifference to the underlying victims: uncertainty about effectiveness, liquidity and future obligations also matter. Critique an inconsistent allocation with those constraints visible rather than read a whole moral psychology from a purchase.

Z · The Craft and the CommunityPurchase Fuzzies and Utilons Separately3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

A motive and an outcome are different variables

To purchase status among your friends

Public recognition can motivate self-serving display, but it can also recruit others or make an effective practice visible. Private donations can be efficient and can still be poorly chosen. Evaluate causal outcomes and incentive effects instead of assuming that sorting acts by psychological motive settles their social value.

A2Evidence gap

The recommended ratio is a heuristic without a fitted model

at least, say, 20 times

A numerical split between feeling good and doing good needs evidence about the person’s motivation, budget and available opportunities. No optimization here derives this ratio or shows it will preserve long-term effort. Present it as a tentative personal experiment, with revision criteria, rather than as the numerical consequence of rational altruism.

A3Logic

There is no general theorem that unbundling is optimal

when you buy separately, optimizing for

Separating emotional reward from impact can clarify a decision. Complementarity can also make a bundle dominate: an activity that sustains motivation may enable donations that otherwise would not occur. Multi-objective optimization does not establish separability; that requires assumptions about costs and interactions that the essay has not demonstrated.

Z · The Craft and the CommunityBystander Apathy4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Individual inhibition and the chance of any help are different outcomes

is that larger groups are less

Philpot and colleagues observed 219 public conflicts across three countries; about nine in ten received help, and more bystanders were associated with a greater chance that someone intervened. This observational, aggregate outcome is not a direct replication of the smoke experiment. It blocks the extrapolation that larger real-world groups necessarily leave a victim less likely to receive any help.

Philpot et al. (2020; online 2019)

A2Scope

The classic laboratory result has genuine but bounded support

smoke start coming up from under the door

Fischer and colleagues’ 2011 meta-analysis pooled 105 effects and over 7,700 participants, finding average individual-level inhibition of helping, g=−0.35. The effect varied with danger and other situational features. A skeptical review should preserve this evidence instead of indiscriminately declaring the bystander effect a failed-replication myth.

Fischer et al. (2011)

A3Survives

Withdrawing an attractive explanation is good practice

is a plausible evolutionary explanation. More

The author’s rejection of his initial evolutionary explanation models a useful correction: a plausible story can fail against the actual experimental design. Its replacement should receive the same scrutiny. Showing one story wrong does not automatically verify the next mechanism proposed.

A4Scope

Teaching a bias is not established immunity to it

telling people about the bystander effect

A demonstration that warning some participants improves helping in a particular setting would be encouraging. It would not establish lasting immunity across emergencies, cultures or online groups. The book’s practical transfer claim needs retention and behavioral tests beyond the original classroom or laboratory context.

Z · The Craft and the CommunityCollective Apathy and the Internet3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

A laboratory mechanism does not automatically explain an online population

instincts, over the Internet?

Online settings differ in visibility, responsibility, coordination costs, duplicate effort and how anyone knows that help occurred. A bystander mechanism is a testable candidate among these possibilities. The essay offers no direct estimate identifying it as the dominant cause of internet inaction.

A2Evidence gap

The reward mechanism could optimize its own display

video of someone asking

Thanks, visible acknowledgment and referral credit might motivate useful work. They can also reward actions that are easy to display, encourage gaming or privilege well-connected participants. Test verified downstream benefit, duplication and cost before treating more visible activity as more effective help.

A3Logic

A successful counterexample changes the question

old statistic about Wikipedia representing 1/2,000 of

Acknowledging an online cooperative success prevents the argument from establishing a general inability to act together. The useful question becomes which infrastructure, incentives and task properties support action. Compare successes with failures instead of using only the latter to diagnose a universal psychological defect.

Z · The Craft and the CommunityIncremental Progress and the Valley3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

An ideal agent is optimal relative to a specified problem

optimality theorems that

An optimality claim needs a model, utilities, available actions and information. It does not guarantee a favorable outcome in a hostile world or identify the right real-world model. Conflating formal optimality with success makes practical approximations look defective for failures no feasible policy could avoid.

A2Survives

Nonmonotonic progress is possible; the far peak is unproven

out of the valley and higher than

Interacting changes can make a partial implementation worse before a complete one becomes better. That logical possibility is real. It does not demonstrate that a specific struggling curriculum sits in such a valley; predeclared milestones, time limits and outside comparisons are needed to distinguish temporary costs from an ineffective method.

A3Logic

A contested application cannot define rationality by agreement

signed up for cryonics, a sudden

Cryonics involves empirical uncertainties and value judgments not settled by identifying familiar biases. Treating uptake as a diagnostic of rationality assumes the conclusion of that practical argument. A valid test should assess the quality of reasoning and calibration without awarding correctness merely for agreement with the author.

Z · The Craft and the CommunityBayesians vs. Barbarians4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Inspectability enables some cooperation under specified assumptions

of each other’s source code. Cooperation in

Program-equilibrium work demonstrates formal ways for agents to condition behavior on each other’s programs. It does not show that every equilibrium is efficient or that one favored decision theory uniquely supplies cooperation. The assumptions about inspection, commitment and computation are doing substantial work.

Tennenholtz (2004)

A2Survives

Rationality does not entail selfish values

utility functions don’t need to be solipsistic

Expected-utility reasoning can represent concern for others, shared projects or self-sacrifice. The essay is right to resist defining rationality as crude material self-interest. The substantive dispute then concerns the objectives and institutional consequences, which the formalism does not select.

A3Scope

Authority remains a hypothesis about comparative performance

If there is no proven method better than a captain

Delegation can be rational without making a leader infallible. A real arrangement needs a way to detect error, replace leadership and distinguish urgent coordination from a convenient excuse to silence correction. Catastrophic stakes increase the importance of those questions.

A4Logic

Stipulated consent does not validate a real coercive institution

survival is to shoot soldiers who disobey

A fictional group’s unanimous advance agreement can be built into the example. Whether a real coercive rule improves survival still depends on competence, mistakes, incentives, abuse and available alternatives. The necessity of coordination supplies no automatic warrant for the particular punishment; the story’s assumptions must not silently become a real-world authorization.

Z · The Craft and the CommunityBeware of Other-Optimizing4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

One apparent success after many attempts still needs retesting

you read the twentieth piece of advice—or

Trying numerous techniques creates opportunities for coincidence, regression to the mean and selective memory. A personally useful observation can justify a further trial without identifying a stable mechanism or predicting another person’s response. Record repeated outcomes and relevant changes in circumstances before declaring the method discovered.

A2Logic

The equality claim is too strong

blog post somewhere; no more and no less likely to work for

Advice from a friend who knows someone’s constraints can contain relevant information that a generic blog post lacks. It can also be overconfident and mistaken, as the anecdote illustrates. One failure establishes neither exact equality of success probabilities nor the uselessness of individualized knowledge.

A3Survives

Respecting refusal is one of the least damaged recommendations

be willing to take “No” for an answer, if you

Individual differences, imperfect knowledge and unequal power give a strong reason to accept that advice may not fit the recipient. This practical and ethical argument does not depend on one priming result. Apply it to rationality teachers and organizers as readily as to intrusive friends.

A4Replication

The relevant power result needs a replication warning

been shown to decrease our ability to empathize with and understand

The post does not identify its study. In the relevant Galinsky et al. 2006 research, Study 2a used power recall and a sarcastic-email perspective task. Many Labs 3 directly replicated that experiment with N=2,969 and found d=0.03, 95% CI [−0.04, 0.10]. This challenges that manipulation and task; it does not establish that power never impairs empathy or judgment.

Ebersole et al. (2016)

Z · The Craft and the CommunityPractical Advice Backed by Deep Theories6 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Evidence gap

An elegant self-experiment does not establish a general treatment effect

Shangri-La Diet per se

A personal dietary result can generate a hypothesis while leaving confounding, adherence and generalizability unresolved. The essay admits that a method may fail for a particular reader, which is appropriate. It still needs controlled evidence for the proposed mechanism and the practical prediction; a deeper story alone does not supply it.

A2Survives

Mechanistic understanding can improve a prediction, if it earns its place

without the explicit theory backing it up.

A theory that correctly predicts when a technique works and fails can improve generalization. A story added after the outcome can explain anything and predict little. The useful test is a new, risky prediction checked against data, including unsuccessful cases and competing explanations.

A3Scope

A memory effect cannot be transferred to meal advice without a test

the Peak–end rule

The classic better-ending experiment concerned cold-water discomfort and later preference between episodes. It does not establish that reducing a meal’s size leaves its remembered pleasure unchanged or that dessert is the only relevant variable. The dietary prescription needs a direct test of the actual outcome being promised.

Kahneman et al. (1993)

A4Replication

The confident appeal to ego depletion aged badly

willpower, mental conflicts, ego depletion, preference reversals,

Hagger et al.’s preregistered 23-lab test found d=0.04, 95% CI [−0.07, 0.15], with 2,141 participants. Vohs et al.’s later 36-lab study found a nonsignificant confirmatory d=0.06 with 3,531 participants. These results substantially weaken reliance on a large, dependable depletion effect as an established foundation for advice.

Hagger et al. (2016); Vohs et al. (2021)

A5Scope

The correction is not that fatigue or every depletion effect is imaginary

a vocabulary of cognitive phenomena that actually exist

A separate preregistered 12-lab study by Dang and colleagues found a small effect, d=0.10, with 1,775 participants. The evidence is heterogeneous, and even a reliable sequential-task effect would not uniquely establish a literal resource tank. The warranted update concerns effect size, generality and mechanism; it does not license an equally sweeping denial.

Dang et al. (2021; online 2020)

A6Logic

A small p-value is not a replication certificate

reported p < 0.01 because p <

The ASA statement rejects treating significance as a universal quality threshold. For illustration, if only 1% of tested hypotheses are real, power is 80% and the false-positive rate is 1%, only about 45% of significant results would be true positives even before selective reporting. These are stipulated numbers, not an estimate for this literature; they show why the threshold alone cannot do the promised work.

Wasserstein & Lazar / ASA (2016)

Z · The Craft and the CommunityThe Sin of Underconfidence4 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Scope

Confidence has several dimensions

confidence, overconfidence, and underconfidence—being interpreted now

Moore and Healy distinguish estimates of one’s performance, placement relative to others and precision of belief. A person can be too pessimistic on one dimension while being too certain on another. The essay’s community-wide diagnosis needs measurements of those dimensions, not just memorable examples of modest people.

Moore & Healy (2008)

A2Evidence gap

A few trials cannot rank your general potential

time out of three

Trying a difficult task can reveal information that introspection misses. A handful of attempts still leaves substantial uncertainty and can be affected by task choice, practice and opponent strength. Treat the result as a local measurement with limits, not a reliable assessment of how high the person can ultimately rise.

A3Logic

Choose a target that does not reward the wrong competence

rationality is systematized winning and trying

A contest can supply concrete feedback, which is better than unmeasured self-assurance. Winning a debate may measure persuasion or opponent weakness more than truth; other challenges likewise reward their scoring rules. Choose an outcome linked to the capacity you want and keep calibration distinct from motivational confidence.

A4Scope

Testing your own strategy survives; a universal first-instinct rule does not

I learned the first form of this

Kruger and colleagues studied the first-instinct fallacy and the tendency to remember costly answer changes disproportionately. Their evidence cautions against a blanket rule to keep the first answer. The essay’s suggestion to examine one’s actual record is the sound component, provided that successes and failures are recorded prospectively.

Kruger, Wirtz & Miller (2005)

Z · The Craft and the CommunityGo Forth and Create the Art!3 notes

Annotations by Astra 6 Pro. Open annotated chapter · Original at reviewed revision

A1Survives

The unfinished art has an identifiable burden of proof

training, teaching, verification

Specify a curriculum and comparison condition, preregister meaningful outcomes, evaluate transfer and retention, report attrition and costs, and allow independent replication. That is a concrete way to pursue the aspiration the ending leaves open. Until such evidence establishes the package’s effect, the Sequences remain a source of hypotheses and arguments, not a demonstrated general technology for becoming stronger.

A2Survives

The invitation to outgrow the text should be taken literally

You should not be getting all your rationality from one author

Reading beyond one author, confronting empirical results and improving the methods are defensible instructions. They authorize the reader to discard unsupported prescriptions and update obsolete examples. The book is strongest when it offers this research agenda rather than presuming that its preferred community already embodies the completed art.

A3Scope

Single authorship is not by itself a cult diagnosis

may have its heroes, it may

One person can originate a valuable framework, while a many-author institution can still be dogmatic. The relevant questions concern deference, independent tests, corrigibility and the consequences of dissent. The warning is useful when it leads to those observations; the number of founding heroes is not a sufficient test.

Evidence you can inspect

Sources and limits of inference.

Original papers and author/publisher records are linked. Years in parentheses distinguish online and issue dates where useful. Effect estimates from different tasks and designs should not be treated as interchangeable.

  1. Fong, Krantz & Nisbett (1986). The effects of statistical training on thinking about everyday problems

    Experiments on statistical reasoning and transfer; not a test of this book or of overall life success.

  2. de Heide & Grünwald (2021). Why optional stopping can be a problem for Bayesians

    Distinguishes Bayesian stopping-rule independence from calibration under different uses of priors.

  3. Sisk et al. (2018). To what extent and under which circumstances are growth mind-sets important to academic achievement? Two meta-analyses

    Small average associations/intervention effects with moderators; not a direct replication of every praise experiment.

  4. Carpenter, Zucker & Avorn (2008). Drug-review deadlines and safety problems

    Observational association between deadline-proximate approvals and safety problems; does not estimate the net mortality effect of regulation.

  5. Bennett (1973). Logical reversibility of computation

    Primary theoretical paper on reversible computation and dissipation.

  6. Halpern (1999). A counterexample to theorems of Cox and Fine

    Limits particular theorem formulations and assumptions; does not refute probability theory.

  7. Gardner (1970). Mathematical Games: The fantastic combinations of John Conway’s new solitaire game “life”

    Reproduction of the original Scientific American description, including births at three neighbors.

  8. Maimonides, Repentance 2:2. Mishneh Torah, Hilchot Teshuvah

    Primary religious text specifying abandonment of wrongdoing and resolve about future conduct.

  9. Gollwitzer & Sheeran (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes

    Evidence about concrete if–then planning, not the efficacy of exhortations to try harder.

  10. Sanchez et al. (2017). Direct replication of Gervais & Norenzayan (2012): No evidence that analytic thinking decreases religious belief

    Direct replication of one acute picture-prime experiment; later related evidence, not a study cited by the essay or a test of multiyear education.

  11. Mihura et al. (2013). The validity of individual Rorschach variables: Systematic reviews and meta-analyses of the Comprehensive System

    Variable-specific validity evidence, with major limitations; does not validate unrestricted inkblot interpretation.

  12. Strupp & Hadley (1979). Specific vs nonspecific factors in psychotherapy: A controlled study of outcome

    Small, selected male college sample and professors chosen for interpersonal ability; not random professors or all psychiatric conditions.

  13. Cuijpers et al. (2023). Cognitive behavior therapy vs. control conditions, other psychotherapies, pharmacotherapies and combined treatment for depression

    409 trials and 52,702 patients; meaningful benefits vs controls, little robust superiority over other psychotherapies.

  14. Morewedge et al. (2015). Debiasing decisions: Improved decision making with a single training intervention

    Experimental improvement on measured biases; bounded evidence of trainability.

  15. Sellier, Scopelliti & Morewedge (2019). Debiasing training improves decision making in the field

    Transfer to a business decision case in graduate training; narrower than general life competence.

  16. Mellers et al. (2014). Psychological strategies for winning a geopolitical forecasting tournament

    Probability training, teamwork and aggregation in scored geopolitical forecasts; not a test of the Sequences curriculum.

  17. Cameron (1999). Raising the stakes in the ultimatum game: Experimental evidence from Indonesia

    Some large offers rejected, alongside greater willingness to accept a given share at higher stakes.

  18. Hill et al. (2011). Co-residence patterns in hunter-gatherer societies show unique human social structure

    32 contemporary foraging societies; indirect evidence about ancestry, with mobility and broad social networks.

  19. Iannaccone (1992). Sacrifice and stigma: Reducing free-riding in cults, communes, and other collectives

    A club-goods model in which costly requirements can support cooperation; not a universal empirical vindication of religion.

  20. Hagger et al. (2016). A multilab preregistered replication of the ego-depletion effect

    23 laboratories; N=2,141; d=0.04, 95% CI [−0.07, 0.15].

  21. Vohs et al. (2021). A multisite preregistered paradigmatic test of the ego-depletion effect

    36 laboratories; N=3,531; nonsignificant confirmatory d=0.06; exploratory analyses differed.

  22. Dang et al. (2021; online 2020). A multilab replication of the ego depletion effect

    12 laboratories; N=1,775; small significant d=0.10, rising to 0.16 after a response-quality exclusion.

  23. Fischer et al. (2011). The bystander-effect: A meta-analytic review on bystander intervention in dangerous and non-dangerous emergencies

    105 independent effect sizes, over 7,700 participants; average individual-level inhibition, moderated by situation.

  24. Philpot et al. (2020; online 2019). Would I be helped? Cross-national CCTV footage shows that intervention is the norm in public conflicts

    219 observed conflicts; at least one helper in roughly 9/10. Aggregate outcome, not a direct replication of laboratory individual inhibition.

  25. Tennenholtz (2004). Program equilibrium

    Formal cooperation through program inspection under specified game assumptions; does not select one practical political order.

  26. Ebersole et al. (2016). Many Labs 3: Evaluating participant pool quality across the academic semester via replication

    Section 4.5 directly replicated Galinsky et al. Study 2a: N=2,969; d=0.03, 95% CI [−0.04, 0.10]. Other power claims require separate assessment.

  27. Wasserstein & Lazar / ASA (2016). The ASA’s statement on p-values: Context, process, and purpose

    Statistical principles: significance thresholds do not establish truth, effect importance or replication probability.

  28. Kahneman et al. (1993). When more pain is preferred to less: Adding a better end

    Cold-water discomfort experiment; not a demonstration that smaller meals produce identical memories.

  29. Moore & Healy (2008). The trouble with overconfidence

    Distinguishes overestimation, overplacement and overprecision; broad labels can hide different directions of error.

  30. Kruger, Wirtz & Miller (2005). Counterfactual thinking and the first instinct fallacy

    Exam-answer changes and remembered regret; evidence against a blanket first-answer rule, not against individual measurement.

  31. Guess & Coppock (2020; online 2018). Does counter-attitudinal information cause backlash? Results from three large survey experiments

    Later boundary evidence, not a direct replication of the exact Taber–Lodge protocol. The 2019 corrigendum corrects Table 1 standard deviations and standard errors: https://doi.org/10.1017/S0007123419000024.

  32. Gieringer (1985), p. 196. The safety and efficacy of new drug approval

    Original decade-based estimate; deaths and casualties differ. Comparison is with foreign regulatory systems, not no regulation.

  33. FDAReview.org, source cited by the essay. Theory, evidence and examples of FDA harm

    Current location of the cited source’s discussion, which reproduces Gieringer’s per-decade figures. The original legacy URL no longer opened in this check.